AI-Ready Data Starts with Observability, Not Governance: says Elton Martins, Data & AI Leader

September 11, 2025

Elton Martins, former data leader at the NFL and Genius Sports, joins Mark Kohout to explore the rising discipline of data observability. From detecting silent failures in data pipelines to enabling AI-ready data ecosystems, Elton shares real-world lessons from scaling data platforms across industries – and explains how observability empowers teams to build trust in their data—before it powers analytics, AI, and decision-making.

• How can data observability organizations detect silent failures before they impact business decisions?
• What’s the difference between data quality management and data observability—and why does it matter for AI readiness?
• What should organizations consider when deciding to build or buy a data observability solution?

Watch the interview

Listen to the podcast

Read the podcast as an interview

(The interview was shortened and edited using ChatGPT)

Mark Kohout: Welcome to the Adastra Podcast. Today we’re talking about data observability with my guest, Elton Martins. Elton is a seasoned data leader, with experience at the NFL and Genius Sports, where he built scalable platforms enabling analytics, real-time insights, and AI use cases. Great to have you here, Elton. What are you working on lately, and what’s keeping you up at night?

Elton Martins: Thanks, Mark. It’s great to be here. Over the past few years, I’ve been laser focused on building data products to power sports analytics—helping leagues and clubs make smarter decisions in marketing, media, and fan engagement. What keeps me up at night is making sure that big investments in modern platforms don’t fall apart because of unreliable data.

Mark Kohout: That’s a common concern. Leaders want ROI from their tech spend but often underestimate the care data needs. Let’s set the stage—what exactly is data observability?

Elton Martins: It’s the ability to monitor and ensure the health of data pipelines. Think of it as practices and tools that catch problems early, before they affect downstream users. The term is new, but the idea comes from DevOps and application monitoring—now applied to data.

Mark Kohout: And why has it become such a priority now?

Elton Martins: A few reasons. First, identification of silent failures: pipelines may look fine, but Finance calls saying numbers don’t match. Second, transparency: regulations and internal stakeholders demand clarity on how data is collected and used. Third, the rise of data mesh and data as a product—teams need autonomy to monitor their domains. And fourth, data drift detection, which is critical for AI/ML models.

Mark Kohout: That’s helpful. But organizations have been doing data quality management for decades. How does observability differ?

Elton Martins: Data quality asks: Is my data fit for purpose? Data observability asks: Is my data flowing and behaving normally? They complement each other, but observability provides real-time visibility into issues that quality frameworks often miss.

Mark Kohout: Can you give me a practical example?

Elton Martins: Sure. If a source usually sends five million records an hour but suddenly only delivers half a million, observability tools flag it instantly. Many use machine learning to learn normal patterns and detect anomalies.

Mark Kohout: So, it helps teams act earlier. What are the main benefits for organizations?

Elton Martins: It reduces time wasted on “fishing expeditions” when issues arise, gives end-to-end lineage, and frees scarce engineering resources. It also improves culture—downstream users aren’t blindsided by data problems the team didn’t catch.

Mark Kohout: What about infrastructure and cost?

Elton Martins: Observability collects metadata like runtime, failure rates, and usage. That helps teams sunset unused assets, identify weak pipeline components, and cut cloud costs, all while improving reliability.

Mark Kohout: And in the AI world, this must be even more important.

Elton Martins: Absolutely. AI requires clean, governed data. Without it, results fall flat, and risks multiply—legal, financial, reputational. In one case, Unity Software fed bad customer data into ML models, which cost them 8% in revenue and a 30% decline in market value.

Mark Kohout: That’s a major impact. From your experience, what advice would you give organizations getting started?

Elton Martins: Start small—focus on mission-critical pipelines, then scale. Don’t treat it as set-and-forget; observability must evolve with your platform. And don’t separate it from governance—they reinforce each other. Finally, decide carefully whether to build or buy. At one company we bought a tool because it fit our stack. At another, we built in-house because our use cases were too unique.

Mark Kohout: That’s great advice. Elton, thank you for sharing your insights and experience. And thanks to our listeners for joining us. Stay tuned for more conversations from Adastra on data, analytics, and AI.

Elton Martins: Thanks, Mark. This was a great discussion.